minor : indentation + assert

examples/gguf/gguf.cpp

@@ -8,6 +8,12 @@
 #include <sstream>
 #include <fstream>
 #include <vector>
+
+#undef MIN
+#undef MAX
+#define MIN(a, b) ((a) < (b) ? (a) : (b))
+#define MAX(a, b) ((a) > (b) ? (a) : (b))
+
 /*
 template<typename T>
 static std::string to_string(const T & val) {
@@ -16,6 +22,7 @@ static std::string to_string(const T & val) {
     return ss.str();
 }
 */
+
 void gguf_ex_write_str(std::ofstream & fout, const std::string & val) {
     const int32_t n = val.size();
     fout.write((const char *) &n, sizeof(n));
@@ -377,11 +384,13 @@ bool gguf_ex_read_2(const std::string & fname) {
 
     struct gguf_file file(fname.c_str(), "rb");
     gguf_mmap data_mmap(&file, 0, false);
+
     const int n_tensors = gguf_get_n_tensors(ctx);
 
     for (int i = 0; i < n_tensors; ++i) {
         const char * name   = gguf_get_tensor_name(ctx, i);
         const size_t offset = gguf_get_data_offset(ctx) + gguf_get_tensor_offset(ctx, i);
+
         struct ggml_tensor * cur = ggml_get_tensor(ctx_data, name);
 
         cur->data = static_cast<char *>(data_mmap.addr) + offset;
@@ -390,11 +399,9 @@ bool gguf_ex_read_2(const std::string & fname) {
         const float * data = (const float *) cur->data;
 
         printf("%s data[:10] : ", name);
-
-        for (int j = 0; j < 10; ++j) {
+        for (int j = 0; j < MIN(10, ggml_nelements(cur)); ++j) {
             printf("%f ", data[j]);
         }
-
         printf("\n\n");
     }
 

gguf-llama.cpp

@@ -508,16 +508,15 @@ struct gguf_load_tensors_map {
 
 enum gguf_file_version {
     GGUF_FILE_VERSION_V1 = 1,
-
 };
 
-
 struct gguf_file_loader {
     gguf_file file;
     gguf_context * gguf_ctx;
     gguf_file_version file_version;
     llama_hparams hparams;
     llama_vocab vocab;
+
     struct ggml_context * ctx_data = NULL;
 
     gguf_file_loader(const char * fname, gguf_load_tensors_map & tensors_map)
@@ -537,7 +536,7 @@ struct ggml_context * ctx_data = NULL;
         read_tensor_metadata(tensors_map);
     }
 
-    uint32_t read_u32(const char * key) {
+    uint32_t read_u32(const char * key) const {
         int i = gguf_find_key(gguf_ctx, key);
         if (i == -1) {
             throw std::runtime_error(format("cannot find param with key %s\n", key));
@@ -546,7 +545,7 @@ struct ggml_context * ctx_data = NULL;
         return gguf_get_val_u32(gguf_ctx, i);
     }
 
-    float read_f32(const char * key) {
+    float read_f32(const char * key) const {
         int i = gguf_find_key(gguf_ctx, key);
         if (i == -1) {
             throw std::runtime_error(format("cannot find param with key %s\n", key));
@@ -555,7 +554,7 @@ struct ggml_context * ctx_data = NULL;
         return gguf_get_val_f32(gguf_ctx, i);
     }
 
-    int read_n_vocab() {
+    int read_n_vocab() const {
         int i = gguf_find_key(gguf_ctx, "tokenizer.ggml.tokens");
         if (i == -1) {
             throw std::runtime_error("cannot find token list in GGUF file\n");
@@ -565,7 +564,6 @@ struct ggml_context * ctx_data = NULL;
     }
 
     void read_hparams() {
-
         // TODO define keys as constants in header
         // TODO: read all hparams from file
 
@@ -606,7 +604,7 @@ struct ggml_context * ctx_data = NULL;
         }
     }
 
-    void read_tensor_metadata(gguf_load_tensors_map & tensors_map) {
+    void read_tensor_metadata(gguf_load_tensors_map & tensors_map) const {
         const int n_tensors = gguf_get_n_tensors(gguf_ctx);
 
         for (int i = 0; i < n_tensors; ++i) {
@@ -614,9 +612,11 @@ struct ggml_context * ctx_data = NULL;
             const char * name = gguf_get_tensor_name(gguf_ctx, i);
 
             struct ggml_tensor * cur = ggml_get_tensor(ctx_data, name);
-            uint32_t n_dims = cur->n_dims;
+
+            const uint32_t n_dims = cur->n_dims;
             tensor.type = cur->type;
             tensor.ne.resize(n_dims);
+
             for (uint32_t j = 0; j < n_dims; ++j) {
                 tensor.ne[j] = cur->ne[j];
             }
@@ -624,6 +624,7 @@ struct ggml_context * ctx_data = NULL;
             if (n_dims < 1 || n_dims > 2) {
                 throw std::runtime_error(format("llama.cpp: tensor '%s' should not be %u-dimensional", name, n_dims));
             }
+
             switch (tensor.type) {
                 case GGML_TYPE_F32:
                 case GGML_TYPE_F16:
@@ -643,7 +644,6 @@ struct ggml_context * ctx_data = NULL;
                 }
             }
 
-
             tensor.file_off = gguf_get_data_offset(gguf_ctx) + gguf_get_tensor_offset(gguf_ctx, i);
 
             tensor.name = name;
@@ -787,6 +787,9 @@ struct gguf_file_saver {
         }
 
         info_offset = file.tell();
+
+        GGML_ASSERT(gguf_get_data_offset(fl->gguf_ctx) >= info_offset);
+
         size_t count = gguf_get_data_offset(fl->gguf_ctx) - info_offset;
         file.write_zeros(count);
         file.seek(info_offset, SEEK_SET);